{
  "recordedAt": "2026-09-08T08:38:28.948294+00:00",
  "kind": "post-hoc-sensitivity-plan",
  "discovery": "The first candidate supplied a verbatim sentence that explicitly denies generalization, but v1 required the literal word inconclusive. This is a genuine verifier false negative.",
  "observedCompletedTrialsAtRevision": [
    "1-candidate",
    "1-baseline",
    "2-baseline",
    "2-candidate"
  ],
  "revision": "verify_v2.py preserves v1 predicates except for three explicitly enumerated, source-grounded equivalent scope sentences. Other failures and unsupported quotations remain failures.",
  "evaluation": "Evaluate every one of the eight retained final answers with both predicates. Preserve timeout/error/incomplete statuses; never substitute a reconstructed result, rerun a model, select only changed outcomes or overwrite original result.json.",
  "controlCoverage": "All six original controls plus three equivalent-scope positives and negative regressions for zero-cost substitution, fabricated quotation suffix and unrelated citation.",
  "inference": "Post hoc sensitivity and verifier diagnosis only. No preregistered success-effect claim; no inference that navigation improves performance.",
  "files": {
    "verify.py": "a528400d84d516370f0f999b9eb6f2f66a8793907ccc4603df92d32f673cbdfd",
    "verify_v2.py": "ca0bacae64bd085f8d60cc5dc54a6834012547606c4483928360e33729068567",
    "test_verifiers.py": "ed237f6e1b301a8cac5a358d90f1ac47108737181439a4c9875048fe01320f6f"
  }
}